Estimating Arterial Wall Deformations from Automatic Key-Point Detection and Matching.

ABSTRACT

Assessing arterial-wall motion and deformations may reveal pathologic alterations in biomechanical properties of the parietal tissues and, thus, contribute to the detection of vascular disease onset. Ultrasound image sequences allow the observation of this motion and many methods have been developed to estimate temporal changes in artery diameter and wall thickness and to track 2-D displacements of selected points. Some methods enable the assessment of shearing or stretching within the wall, but none of them can estimate all these deformations simultaneously. The method herein proposed was devised to simultaneously estimate translation, compression, stretching and shearing of the arterial wall in ultrasound B-mode image sequences representing the carotid artery longitudinal section. Salient blob-like patterns, called key points, are automatically detected in each frame and matched between successive frames. A robust estimator based on an affine transformation model is then used to assess frame-to-frame motion explaining at best the key-point matches and to reject outliers. Realistic simulated image sequences were used to evaluate the accuracy and robustness of the method against ground truth. The method was also visually assessed on clinical image sequences, for which true deformations are unknown.